Brain Regions Responsive to Novelty in the Absence of Awareness

See allHide authors and affiliations

Science  23 May 1997:
Vol. 276, Issue 5316, pp. 1272-1275
DOI: 10.1126/science.276.5316.1272


Brain regions responsive to novelty, without awareness, were mapped in humans by positron emission tomography. Participants performed a simple reaction-time task in which all stimuli were equally likely but, unknown to them, followed a complex sequence. Measures of behavioral performance indicated that participants learned the sequences even though they were unaware of the existence of any order. Once the participants were trained, a subtle and unperceived change in the nature of the sequence resulted in increased blood flow in a network comprising the left premotor area, left anterior cingulate, and right ventral striatum. Blood flow decreases were observed in the right dorsolateral prefrontal and parietal areas. The time course of these changes suggests that the ventral striatum is responsive to novel information, and the right prefrontal area is associated with the maintenance of contextual information, and both processes can occur without awareness.

The detection of novelty is a cognitive operation necessary to survival and requires an assessment of both expectedness and context. Events can be familiar in one context but novel in another. More precisely, novelty represents a deviation from the expected likelihood of an event on the basis of both previous information and internal estimates of conditional probabilities (1). Novelty detection has typically been linked to consciousness because novel events often capture attention. For similar reasons, studies of novelty have often been confounded by awareness (2). Here, we sought to determine whether the response to novelty can occur without awareness and, if so, to identify the associated brain regions in a manner unconfounded by awareness. To do so, we used an implicit learning task.

A large body of research has examined learning mechanisms that operate below the level of awareness. This type of learning is said to occur implicitly because behavioral measures indicate that learning takes place, even though the individuals are unaware of this or are unable to report it explicitly (3). A frequently used paradigm is based on a serial reaction-time task, in which participants observe sequences of visual stimuli and must press buttons corresponding to these. Unknown to the participants, the sequence of stimuli is predetermined by a fixed, repeating order. With practice, reaction times improve (compared with randomly sequenced stimuli), indicating that the participants have learned about the sequential order. However, they are not always conscious of this. When the sequence is sufficiently complex, individuals are unaware of the sequential regularities or that they have learned anything specific about the stimuli, even though their reaction times have improved significantly (4). This indicates that sequential information can be both learned and used in the absence of awareness.

One type of sequence that has been well studied is based on finite-state grammars (5). Such grammars can be used to generate highly complex, context-dependent sequences. With enough practice, individuals show improvements consistent with implicit learning of such grammars. However, because such grammars are typically probabilistic, specific repeating sequences rarely occur, further reducing the likelihood of awareness of the sequential regularities.

Implicit learning of finite-state grammars means that participants have developed expectations for each stimulus, on the basis of the specific stimuli that preceded it in the sequence (that is, its context). Under such conditions, changing the rules of the grammar will cause subsequent stimuli to violate these expectations, by appearing in novel contexts. Thus, a switch in grammars should engage mechanisms sensitive to novelty. Furthermore, if participants are unaware of both the initial grammar and the switch, then changes in behavior and regional brain activity associated with the switch should represent processes occurring without awareness.

Ten normal, right-handed volunteers performed a serial reaction-time task while regional cerebral blood flow (rCBF) was mapped by positron emission tomography (PET) (6). The task consisted of a number (1, 2, or 3) appearing on a computer screen, and the participant was instructed to press the corresponding key on the number pad with the right hand as quickly and as accurately as possible. The sequence of stimuli was determined by a finite-state grammar and appeared every 700 ms in 90-s blocks with 45-s rests between blocks. A total of 53 blocks were performed (7). For the first 28 blocks the sequence of stimuli followed one grammar (grammar A), after which the sequence was switched, without telling the participants, to a second grammar (grammar B), and another 25 blocks were presented (8). PET scans were performed starting with the first block and every fourth block thereafter (9). Fourteen scans were obtained during the course of the task, and participants continued to perform the task between scans. Two additional PET scans were taken during a rest condition (staring at a box on the screen) before the task was begun and after completion of it. Areas of significant difference in rCBF after the grammar change were identified with statistical parametric maps (10).

The median reaction time (RT) for the 10 participants declined from an initial average of 500 ms to 450 ms at the end of grammar A (Fig.1). When the grammar changed, RT increased to 465 ms and then declined again. The average RT increase of 15 ms, although statistically significant, was undetected by the participants, as was the change in grammar (11). After the initial block of stimuli, the accuracy remained stable at 92% and showed no significant change from grammar A to grammar B.

Figure 1

Average median reaction times (RT) ± SE as a function of trial block. The average of the median RTs across participants is shown with separate regression curves that followed a power law for each grammar. The switch from grammar A to grammar B resulted in a 15-ms increase in median RT. There was no significant change in accuracy across the grammar change.

Areas of significantly increased rCBF after the grammar change are shown in Fig. 2 (12). Three areas of focal increase were noted: left premotor area, left anterior cingulate, and right ventral striatum and nucleus accumbens. The scan-by-scan plots of rCBF indicate that the increase in blood flow in these areas was clearly related to the grammar change, with the ventral striatum showing the most significant increase in response to the grammar change. The most prominent areas of decreased rCBF (Fig.3) were in the right dorsolateral prefrontal cortex (Brodman area 45/46) and right inferior parietal and superior temporal areas. These areas were also negatively correlated with RT and displayed a different time course from the other areas (13).

Figure 2

Statistical parametric maps (SPMs) for areas of increased blood flow after the grammar change (upper row) and the corresponding time course of rCBF ±SE for the significant area in each slice (lower row). Contrasts between the end of grammar A and the beginning of grammar B were performed by subtracting the activity maps from the four final scans of grammar A from the four beginning scans of grammar B and performing paired t tests on each voxel. Only differences that were significant at P < 0.001 were considered in the interpretation, but voxels significant atP < 0.01 are shown for demonstration of the extent of the activity differences. The SPMs are superimposed on the average MRI for the cohort of 10 participants. This blood-flow contrast isolates those brain regions showing significant rCBF changes related to changing stimulus probability, which occurred largely without awareness. Areas of significant increase in blood flow included, from left to right [Talairach-Tournoux (TT) coordinates; peak Z values]: left premotor area (−28, 6, 52; Z = 3.10), left anterior cingulate (−4, 38, 12; Z = 3.36), and right ventral striatum and nucleus accumbens (4, 10, −4; Z = 3.40). The ventral striatum displayed a time course consistent with response to novelty.

Figure 3

SPM for areas of blood flow negatively correlated with RT (left). The ventral region extends from the right dorsolateral prefrontal cortex (DLPFC) (Brodman area 45/46; TT coordinates: 36, 20, 16) posteriorly to primary sensory area (TT coordinates: 44, −16, 20). The posterior region is centered on the junction of the right inferior parietal lobule and superior temporal gyrus (TT coordinates: 44, −50, 16). The rCBF time course of the DLPFC (right) is consistent with the encoding and maintenance of contextual information during the first grammar, which declines with the grammar switch and increases again as the second grammar is learned.

Different regions showed different time courses of rCBF during task performance, which can be used to distinguish the functions of the regions. For example, the right cerebellum showed a constant elevation of rCBF during task performance relative to rest, consistent with a role in sensory or motor processes associated with task execution. Other areas, such as left premotor cortex, showed a pattern that varied during each grammar but was similar for both. This could reflect an involvement in the implicit learning of each grammar, with greatest activity early when learning was the greatest.

The striatum exhibited a different pattern: an increase in rCBF early during each grammar that lasted significantly and substantially longer during the second grammar. This pattern is consistent with a response to novelty. This area was transiently activated at the beginning of the first grammar, when the task itself was novel; however, the basic task was simple and easily learned, so this was a brief response. With practice, stimuli came to be associated with the particular sequential contingencies of the first grammar. When the grammar was switched, these contingencies were violated. This violation of the learned contingencies by the stimuli was a more powerful elicitor of novelty than their appearance at the very beginning of the experiment, when they were not associated with any specific expectations. Previous brain mapping studies have pointed to the role of the right ventral striatum in implicit sequence learning (14); however, the time course of the rCBF observed in our experiment, particularly its strong response to the grammar switch, support a more specific response to novelty (15). This sensitivity to novelty may also explain the known dopaminergic response to errors in reward prediction (16). The striatum is reciprocally connected to the dopamine neurons of both the substantia nigra and ventral tegmentum, so the striatal response could signal prediction errors of reinforcing stimuli.

Our findings identified regions involved in processing novelty of a particular type. An event can be novel because the stimulus has never been seen before, or because it is being seen in a new or unexpected context. Furthermore, individuals may or may not be aware of the novelty of a stimulus. Previous imaging studies of novelty detection have focused primarily on the conscious processing of unfamiliar stimuli. These studies found activation of a number of limbic structures, including hippocampus, parahippocampal gyrus, medial dorsal thalamus, and medial frontal cortex (2), that differ from the regions identified in our experiment. These differences may be due to the fact that our experiment involved familiar rather than unfamiliar stimuli, and unconscious rather than conscious detection of novelty. Thus, the structures we identified could participate in a system specialized for the detection of familiar stimuli that violate expectations, novelty in the absence of awareness, or both.

A different profile of rCBF was observed in the right dorsolateral prefrontal cortex (DLPFC). This region showed a progressive increase in rCBF during the first grammar, with a decrease after the grammar switch and then an increase late during the second grammar. Several studies have implicated DLPFC in the active maintenance of task-relevant information (17) and, in particular, the maintenance of context information (18). Superficially, our task did not require the maintenance of such information. Participants simply responded to the current stimulus. However, the improvements in RT indicate that they were learning sequential dependencies among the stimuli and using these to anticipate the next stimulus. With practice, participants both learned and maintained longer, and more predictive, sequences.

The changes in rCBF that occurred in right DLPFC during the first grammar are consistent with the operation of maintenance mechanisms. The progressive increase in rCBF could reflect the increase in maintenance requirements associated with longer sequences as these were learned and used (19). This is consistent with studies demonstrating that activation of a similar region of DLPFC increases progressively as memory load for sequential information is increased (20). Blood flow within this region initially decreased during the second grammar, presumably because the sequences learned during the first were no longer informative. Eventually, rCBF began to increase again, as the new sequences were learned and used.

Our findings also address an important question regarding the relation of active maintenance, prefrontal function, and awareness. Previous studies implicating prefrontal cortex in active maintenance have involved circumstances in which individuals were aware of the need to maintain information. This is consistent with the belief that active maintenance requires conscious mental effort. Our findings call this idea into question. The participants in our study were not aware of either the grammars or the switch and therefore had no conscious reason to maintain information about prior stimuli. Therefore, our findings suggest that the prefrontal mechanisms involved in active maintenance, like those responsive to novelty, can operate independently of awareness.

In summary, we have identified structures associated with performance in an implicit learning task that, by their profile of activity, can be associated with specific cognitive functions. Our findings are consistent with the view that DLPFC is involved in the active maintenance of context information used for prediction, and that the ventral striatum monitors the reliability of such predictions, becoming activated when these are violated by stimuli that appear in an unexpected context. We found that both mechanisms can operate without awareness.


View Abstract

Navigate This Article